"""Build a shareable `shares/` folder from the EDGAR research DB.

Exports research-ready datasets (parquet + CSV), metadata, diagnostics,
the PDF report and figures, plus a text sample — everything small enough
to share without the 88 GB of raw filings.
"""
from pathlib import Path
import shutil
import json

import duckdb

ROOT = Path(__file__).resolve().parent.parent
SHARES = ROOT / "shares"

PARQUET = ROOT / "data" / "parquet"

# (source parquet dir, output base name, friendly description)
DATASETS = [
    ("annual_final", "annual_panel_final", "Annual firm-year panel WITH engineered features (deltas, rolling, novelty)"),
    ("quarterly_final", "quarterly_panel_final", "Quarterly firm-quarter panel with sector labels"),
    ("filing_scores", "filing_scores", "One row per individual filing with 8 research-construct scores"),
    ("filings", "filings_index", "Master index of every filing downloaded"),
    ("companies", "companies", "Company reference table: S&P 500 ∪ Russell 3000 (CIK, ticker, sector, SIC, index flags)"),
]


def reset_dir(p: Path):
    if p.exists():
        shutil.rmtree(p)
    p.mkdir(parents=True, exist_ok=True)


def main():
    reset_dir(SHARES)
    datasets_dir = SHARES / "datasets"
    meta_dir = SHARES / "metadata"
    diag_dir = SHARES / "diagnostics"
    report_dir = SHARES / "report"
    samples_dir = SHARES / "samples"
    for d in (datasets_dir, meta_dir, diag_dir, report_dir, samples_dir):
        d.mkdir(parents=True, exist_ok=True)

    con = duckdb.connect()
    manifest = []

    # 1. Research-ready datasets -> parquet + csv
    for src, out, desc in DATASETS:
        glob = f"{PARQUET / src}/**/*.parquet"
        df = con.execute(f"SELECT * FROM read_parquet('{glob}')").df()
        pq = datasets_dir / f"{out}.parquet"
        csv = datasets_dir / f"{out}.csv"
        df.to_parquet(pq, index=False, compression="zstd")
        df.to_csv(csv, index=False)
        manifest.append({"name": out, "rows": len(df), "cols": df.shape[1], "desc": desc})
        print(f"  [dataset] {out:24s} {len(df):>7,} rows x {df.shape[1]:>3} cols")

    # 2. Metadata JSON (constituents + cik mapping; skip the noisy download log)
    for fn in ("universe.json", "sp500_constituents.json", "russell3000_constituents.json", "cik_mapping.json"):
        s = ROOT / "data" / "metadata" / fn
        if s.exists():
            shutil.copy2(s, meta_dir / fn)
            print(f"  [metadata] {fn}")

    # 3. Diagnostics (CSV summaries + coverage report)
    for f in sorted((ROOT / "outputs" / "diagnostics").glob("*")):
        if f.suffix in (".csv", ".json", ".txt"):
            shutil.copy2(f, diag_dir / f.name)
    print(f"  [diagnostics] {len(list(diag_dir.iterdir()))} files")

    # 4. Report PDFs + figures
    latex = ROOT / "outputs" / "latex"
    for pdf in ("edgar_research_report.pdf", "research_questions.pdf"):
        if (latex / pdf).exists():
            shutil.copy2(latex / pdf, report_dir / pdf)
    fig_src = latex / "figures"
    if fig_src.exists():
        shutil.copytree(fig_src, report_dir / "figures", dirs_exist_ok=True)
    print(f"  [report] 2 PDFs + figures")

    # 5. Text sample so reviewers see the underlying parsed content (500 rows)
    sec_glob = f"{PARQUET / 'filing_sections'}/**/*.parquet"
    sample = con.execute(
        f"SELECT accession_number, section_name, item_number, word_count, "
        f"LEFT(text, 1500) AS text_preview "
        f"FROM read_parquet('{sec_glob}') USING SAMPLE 500 ROWS"
    ).df()
    sample.to_csv(samples_dir / "filing_sections_sample.csv", index=False)
    print(f"  [sample] filing_sections_sample.csv ({len(sample)} rows)")

    # write a machine-readable manifest
    (SHARES / "manifest.json").write_text(json.dumps(manifest, indent=2))
    print("\nDone -> shares/")


if __name__ == "__main__":
    main()